Baltimore Life Expectancy

The purpose of this project is to predict life expectancy in Baltimore on a single block resolution. The main source of data for life expectancy at the neighborhood level are the Baltimore City Health Department Neighborhood Health Profiles. I plan on utilizing free data provided by OpenBaltimore, which appears to have a more cleaned up version of the above data set in the Children and Family Health & Well-Being (2010-2014) and Housing and Community Development (2010-2014) reports. The codebook is here.

Main question

Can we predict life expectancy at the block-level resolution in Baltimore using various health, crime, and economic statistics collected at the individual house level and/or aggregated to the block or neighborhood level?

Datasets that I have downloaded

• Children and Family Health & Well-Being (2010-2014)
• Housing and Community Development (2010-2014)
• Building vacancies in Baltimore which provides the street and neighborhood of all current building vacancies. This is updated at least twice a month
• Victim based crime. Accessing this on Sept. 13, 2016 the most recent crime listed occurred on Sept. 3, 2016
• Real property taxes at an address level
• Liquor licenses
• 2010 Census data at a neighborhood level
• Grocery stores. Contains type of supermarket and location. Could be interesting to look at distance to closest grocery store, and clusters of small or limited groceries stores predict health outcomes
• BPD Arrests. Includes address of arrest for some, but not all arrests
• Baltimore prositution
• Block-level state shape file
• Restaurants in Baltimore
• [ACS Block Group Household Income Data]

Preprocessing

• Have used codebook available online to rename variables so they are more easily understandable
• Have cleaned up latitude and longitude where it applies
• Lots of missing locations, especially on arrest data
  • Have to investigate if this is random throughout police districts, or is biased
• Use Google Maps API to get latitude and longitude for any data sets without this. Then use this to infer neighborhood & block for any datasets where neighborhood is not listed
  • Use R package rgdl to map property taxes by latitude and longitude to csa
• Clean up dates
• “Tidy” data. Separate data sets with numeric and character variables. Use either neighborhood or block as key

Visualizations

Analysis plan

• Continue to identify any data sets/variables that need cleaning
• Read downsampling literature
• Decide whether to take a “machine learning approach” or more informed statistical approach
  • By “informed” statistical approach, that means I do some variable selection on my own and use more traditional technique like LM/GLM
• To identify variables of interest, there are two things to pay attention to:
  • How homogeneous they are within neighborhood AND potentially correlated with life expectancy
  • They can be aggregated at CSA scale and can be predicted from house/block level data
• Decide how to use neighborhood spatial correlation
• Apply method to predict at CENSUS block level

Packages

• maps
• maptools
• rgdal
• rgeos
• ggmap
• tigris